@pome-sh/cli 0.23.31 → 0.23.33

package/dist/build-info.json

@@ -1,6 +1,6 @@
 {
   "package": "pome-sh",
-  "version": "0.23.31",
-  "git_sha": "1352a9855e65e27f1717eb267aa1b6bec404dc96",
-  "build_time": "2026-08-13T00:49:39.745Z"
+  "version": "0.23.33",
+  "git_sha": "224f20c230aa9a79b6e074916240940da903dcaf",
+  "build_time": "2026-08-13T01:09:10.392Z"
 }

package/dist/{checks-QMHVQKS7.js → checks-PZSMAVLY.js}

@@ -183,7 +183,7 @@ async function checkTwinReachable(_configDir) {
   });
   let harness;
   try {
-    const { bootTwin } = await import('./twinHarness-VX3GE7QB.js');
+    const { bootTwin } = await import('./twinHarness-XO47SYBH.js');
     harness = await bootTwin({
       twin: "github",
       seedState: void 0,

package/dist/{chunk-5JJDFIGJ.js → chunk-3OVV6JUQ.js}

@@ -128,9 +128,9 @@ var TWIN_REGISTRY = {
     portEnvName: "GMAIL_TWIN_PORT",
     tokenEnvName: "POME_GMAIL_TOKEN",
     version: package_default4.version,
-    defaultSeed: async () => (await import('./src-7F4QNBTF.js')).defaultSeedState(),
+    defaultSeed: async () => (await import('./src-A6EH5KVF.js')).defaultSeedState(),
     async boot({ seedState, runId, recorder }) {
-      const { createGmailTwinApp, GmailDomain, openGmailTwinDatabase, parseSeed } = await import('./src-7F4QNBTF.js');
+      const { createGmailTwinApp, GmailDomain, openGmailTwinDatabase, parseSeed } = await import('./src-A6EH5KVF.js');
       const db = openGmailTwinDatabase(":memory:");
       const seed = parseSeed(seedState);
       const domain = new GmailDomain(db);

package/dist/{chunk-EKOZ32F2.js → chunk-A2LGBYJG.js}

@@ -1,5 +1,5 @@
 import { readManifest, normalizeManifestTwins } from './chunk-ABM3CMQB.js';
-import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-OOYWFDEJ.js';
+import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-ZTINFLSV.js';
 import { MOUNTED_TWINS, HostedAuthError, HostedDiscardRefusedError, HostedQuotaError, HostedOrchError, HostedTrialError, agentResponseSchema } from './chunk-X66JOOO7.js';
 import { redactSecrets, redactEvent } from './chunk-SG6ZTIMT.js';
 import { existsSync } from 'node:fs';

package/dist/{chunk-66DVJZMX.js → chunk-GEC6MJV5.js}

@@ -1,7 +1,7 @@
 import { getAvailablePort } from './chunk-XDU6TD4O.js';
 import { buildEgressAllowlist, readBlockedEgress } from './chunk-CBFKZZBR.js';
-import { parseTaskFile, seedStateForTwin, runAgentCommand, writeRunArtifactsCore } from './chunk-OOYWFDEJ.js';
-import { createRecorder, bootTwin } from './chunk-A46IA7EK.js';
+import { parseTaskFile, seedStateForTwin, runAgentCommand, writeRunArtifactsCore } from './chunk-ZTINFLSV.js';
+import { createRecorder, bootTwin } from './chunk-YL4FHI3J.js';
 import { eventSchema } from './chunk-VBATFCWR.js';
 import { redactSecrets, redactEvent } from './chunk-SG6ZTIMT.js';
 import { serve } from '@hono/node-server';

package/dist/{chunk-SE44FKWH.js → chunk-KTVFRR56.js}

@@ -429,10 +429,10 @@ var mailboxLabelCount = defineCheck({
   },
   evaluate({ mailbox, count, label }, { final }) {
     const wanted = Number(count);
-    if (final.messages == null || final.messageLabels == null) {
+    if (final.messages == null || final.messageLabels == null || final.labels == null) {
       return { passed: false, status: "skipped", reason: "state_incomplete" };
     }
-    if (isTruncated(final, "messages") || isTruncated(final, "messageLabels")) {
+    if (isTruncated(final, "messages") || isTruncated(final, "messageLabels") || isTruncated(final, "labels")) {
       return { passed: false, status: "skipped", reason: "collection_truncated" };
     }
     if (final.mailboxes != null && !final.mailboxes.some((mb) => (mb.email ?? "").toLowerCase() === mailbox.toLowerCase())) {
@@ -489,10 +489,10 @@ var oneMessagePerRecipient = defineCheck({
   },
   evaluate({ label, count }, { final }) {
     const wanted = parseCount(count);
-    if (final.messages == null || final.messageLabels == null) {
+    if (final.messages == null || final.messageLabels == null || final.labels == null) {
       return { passed: false, status: "skipped", reason: "state_incomplete" };
     }
-    if (isTruncated(final, "messages") || isTruncated(final, "messageLabels")) {
+    if (isTruncated(final, "messages") || isTruncated(final, "messageLabels") || isTruncated(final, "labels")) {
       return { passed: false, status: "skipped", reason: "collection_truncated" };
     }
     const ids = labelIdsFor(final, label);

package/dist/{chunk-A46IA7EK.js → chunk-YL4FHI3J.js}

@@ -1,4 +1,4 @@
-import { TWIN_NAMES, isTwinName, TWIN_REGISTRY } from './chunk-5JJDFIGJ.js';
+import { TWIN_NAMES, isTwinName, TWIN_REGISTRY } from './chunk-3OVV6JUQ.js';
 import { createFileBackedRecorderStore, createRecorderStore } from './chunk-OW7VVW6X.js';
 
 // src/recorder/recorder.ts

package/dist/{chunk-OOYWFDEJ.js → chunk-ZTINFLSV.js}

@@ -2,7 +2,7 @@ import { criterionSchema, finalizeResponseSchema, HostedDiscardRefusedError, Hos
 import { seedSchema, parseSeed, defaultSeedState as defaultSeedState$2 } from './chunk-KIUIKAVU.js';
 import { gmailSeedSchema, defaultSeedState } from './chunk-NJ246QPJ.js';
 import { linearSeedSchema, defaultSeedState as defaultSeedState$1 } from './chunk-ZKID2HS3.js';
-import { isTwinName, TWIN_REGISTRY } from './chunk-5JJDFIGJ.js';
+import { isTwinName, TWIN_REGISTRY } from './chunk-3OVV6JUQ.js';
 import { toTwinHttpEventRow } from './chunk-OW7VVW6X.js';
 import { redactEvent, redactSecrets } from './chunk-SG6ZTIMT.js';
 import { mkdir, appendFile, writeFile, readFile } from 'node:fs/promises';

package/dist/{runDemo-6WVLS72G.js → runDemo-EV3CQBDR.js}

@@ -1,16 +1,16 @@
 import { newGroupId, reassuranceBox, twinReadyLine, trialsHeaderLine, trialLine, summaryLines, evaluatingLine, criterionPhrase } from './chunk-RGZBC7NF.js';
 import { DemoCapacityError, capacityLabel, parseCapacityMarker, capacityKindFrom } from './chunk-ZX4WNSZ5.js';
-import { runTask, demoTaskPath, DEMO_TASK_NAME, DEMO_REPO } from './chunk-66DVJZMX.js';
+import { runTask, demoTaskPath, DEMO_TASK_NAME, DEMO_REPO } from './chunk-GEC6MJV5.js';
 import { getAvailablePort } from './chunk-XDU6TD4O.js';
 import './chunk-CBFKZZBR.js';
-import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, outcomeOf } from './chunk-OOYWFDEJ.js';
+import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, outcomeOf } from './chunk-ZTINFLSV.js';
 import './chunk-NW7HGA2K.js';
 import { HostedQuotaError, HostedOrchError } from './chunk-X66JOOO7.js';
 import './chunk-KIUIKAVU.js';
 import './chunk-NJ246QPJ.js';
 import './chunk-ZKID2HS3.js';
-import { bootTwin } from './chunk-A46IA7EK.js';
-import './chunk-5JJDFIGJ.js';
+import { bootTwin } from './chunk-YL4FHI3J.js';
+import './chunk-3OVV6JUQ.js';
 import './chunk-OW7VVW6X.js';
 import './chunk-VBATFCWR.js';
 import './chunk-SG6ZTIMT.js';

package/dist/{runTrialGroup-BMQ2DPSL.js → runTrialGroup-WXDRCVUE.js}

@@ -1,13 +1,13 @@
 import { newGroupId, criterionPhrase } from './chunk-RGZBC7NF.js';
-import { runTaskHosted, resolveRunAgentIdentity } from './chunk-EKOZ32F2.js';
+import { runTaskHosted, resolveRunAgentIdentity } from './chunk-A2LGBYJG.js';
 import './chunk-ABM3CMQB.js';
-import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-OOYWFDEJ.js';
+import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-ZTINFLSV.js';
 import './chunk-NW7HGA2K.js';
 import { HostedQuotaError, HostedTrialError } from './chunk-X66JOOO7.js';
 import './chunk-KIUIKAVU.js';
 import './chunk-NJ246QPJ.js';
 import './chunk-ZKID2HS3.js';
-import './chunk-5JJDFIGJ.js';
+import './chunk-3OVV6JUQ.js';
 import './chunk-OW7VVW6X.js';
 import './chunk-VBATFCWR.js';
 import './chunk-SG6ZTIMT.js';

package/dist/src/cli/main.js

@@ -1,22 +1,22 @@
 #!/usr/bin/env node
-import { DEFAULT_CONTROL_PLANE_URL, DEFAULT_DASHBOARD_URL, clearLocalCredentials, friendlyHostedError, runSessionCreate, runSessionList, runSessionStop, resolveCredentials, runTaskHosted, discoverRunSet, VERDICT_ARTIFACT_VERSION, persistCredentialsAfterLogin, DEFAULT_DOCS_SITE_ORIGIN, resolveSeams, resolveCachedAgentId, readLinkCache, postAgentResolver, writeLinkCache, ensurePomeGitignored } from '../../chunk-EKOZ32F2.js';
+import { DEFAULT_CONTROL_PLANE_URL, DEFAULT_DASHBOARD_URL, clearLocalCredentials, friendlyHostedError, runSessionCreate, runSessionList, runSessionStop, resolveCredentials, runTaskHosted, discoverRunSet, VERDICT_ARTIFACT_VERSION, persistCredentialsAfterLogin, DEFAULT_DOCS_SITE_ORIGIN, resolveSeams, resolveCachedAgentId, readLinkCache, postAgentResolver, writeLinkCache, ensurePomeGitignored } from '../../chunk-A2LGBYJG.js';
 import { readManifest, writeManifest, MANIFEST_JSON, readRequiredManifest, normalizeManifestTwins } from '../../chunk-ABM3CMQB.js';
-import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-66DVJZMX.js';
+import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-GEC6MJV5.js';
 import '../../chunk-XDU6TD4O.js';
 import '../../chunk-CBFKZZBR.js';
-import { parseTaskFile, scoreStatus, runScoreLine, readLatestRun, readMetaSummary, outcomeOf, readConfigTwins, scoreCountsSummary, markerFor, criterionMarkerLabel, twinSkipSuffix, readCodeCriteria, createHostedClient, toTwinHttpEvent, redactJsonl, scoreFromFinalizeResponse, parseGitHubSeedState, uploadRunBlobs, isPreSatisfied } from '../../chunk-OOYWFDEJ.js';
+import { parseTaskFile, scoreStatus, runScoreLine, readLatestRun, readMetaSummary, outcomeOf, readConfigTwins, scoreCountsSummary, markerFor, criterionMarkerLabel, twinSkipSuffix, readCodeCriteria, createHostedClient, toTwinHttpEvent, redactJsonl, scoreFromFinalizeResponse, parseGitHubSeedState, uploadRunBlobs, isPreSatisfied } from '../../chunk-ZTINFLSV.js';
 import '../../chunk-NW7HGA2K.js';
 import { MOUNTED_TWINS, deriveAgentSlug, exitCodeFor, HostedUsageError, HostedOrchError, HostedAuthError, HostedQuotaError } from '../../chunk-X66JOOO7.js';
 import { TAPE_ASSERTABLE_TOOLS } from '../../chunk-FKZZWWYC.js';
 import { seedSchema } from '../../chunk-KIUIKAVU.js';
 import { SLACK_CHECKS } from '../../chunk-IDNSKKEC.js';
-import { GMAIL_CHECKS } from '../../chunk-SE44FKWH.js';
+import { GMAIL_CHECKS } from '../../chunk-KTVFRR56.js';
 import '../../chunk-NJ246QPJ.js';
 import { LINEAR_CHECKS } from '../../chunk-XBT6ZLBG.js';
 import '../../chunk-ZKID2HS3.js';
 import { oneOf, defineCheck, repoRef, VACUITY_SENTINEL_NUMBER, childStatePath, VACUITY_SENTINEL, statePath, templateSlots, renderCheck, checksDigest, checkPattern, checkNearMissPattern } from '../../chunk-JWJYNAWI.js';
-import '../../chunk-A46IA7EK.js';
-import { TWIN_NAME_LIST, isTwinName, createGitHubSmokeApp } from '../../chunk-5JJDFIGJ.js';
+import '../../chunk-YL4FHI3J.js';
+import { TWIN_NAME_LIST, isTwinName, createGitHubSmokeApp } from '../../chunk-3OVV6JUQ.js';
 import '../../chunk-OW7VVW6X.js';
 import { isLegacyEventRow, eventSchema } from '../../chunk-VBATFCWR.js';
 import { redactEvent, redactSecrets } from '../../chunk-SG6ZTIMT.js';
@@ -3064,7 +3064,22 @@ async function runChecksCommand(twinArg, opts) {
 // src/task/insertCriterion.ts
 var HEADING_RE = /^##\s+Success Criteria\s*$/;
 var NEXT_HEADING_RE = /^##\s+/;
-var CRITERION_RE = /^[-*]\s+\[(?:code|model)(?::[a-z][a-z0-9_-]*)?(?:\s+always-scored)?\]\s+.+$/;
+var CRITERION_RE = /^[-*]\s+\[(code|model)(?::([a-z][a-z0-9_-]*))?(\s+always-scored)?\]\s+(.+)$/;
+function identityKey(criterion) {
+  return `${criterion.kind}\0${criterion.twin ?? ""}\0${criterion.text}`;
+}
+function readCriterionLine(line) {
+  const match = line.trim().match(CRITERION_RE);
+  if (!match) return void 0;
+  return { kind: match[1], tag: match[2], text: match[4].trim() };
+}
+function primaryTwin(source) {
+  try {
+    return readConfigTwins(source)[0];
+  } catch {
+    return void 0;
+  }
+}
 var MissingCriteriaSectionError = class extends Error {
   constructor(path) {
     super(
@@ -3074,11 +3089,15 @@ var MissingCriteriaSectionError = class extends Error {
   }
 };
 var DuplicateCriterionError = class extends Error {
-  constructor(path, line) {
+  constructor(path, line, existing = line) {
+    const differs = existing.trim() !== line.trim();
     super(
       `${path} already has this criterion:
-  ${line}
-Adding it again would score it twice and inflate the task's denominator.`
+  ${existing.trim()}
+` + (differs ? `which is the same check as the one being added:
+  ${line.trim()}
+\u2014 a marker annotation or a twin tag does not make it a second check.
+` : "") + `Adding it again would score it twice and inflate the task's denominator.`
     );
     this.name = "DuplicateCriterionError";
   }
@@ -3094,13 +3113,33 @@ function insertCriterion(source, line, path = "this task file") {
       break;
     }
   }
+  const primary = primaryTwin(source);
+  const declared = /* @__PURE__ */ new Map();
+  for (const criterion of readCodeCriteria(source)) {
+    declared.set(
+      identityKey({ kind: "code", twin: criterion.twin, text: criterion.text }),
+      `- ${criterion.marker} ${criterion.text}`
+    );
+  }
   let insertAt = -1;
   for (let i = start + 1; i < end; i += 1) {
-    const existing = lines[i].trim();
-    if (!CRITERION_RE.test(existing)) continue;
-    if (existing === line.trim()) throw new DuplicateCriterionError(path, line);
+    const parsed = readCriterionLine(lines[i]);
+    if (!parsed) continue;
+    if (parsed.kind === "model") {
+      declared.set(
+        identityKey({ kind: "model", twin: parsed.tag ?? primary, text: parsed.text }),
+        lines[i].trim()
+      );
+    }
     insertAt = i + 1;
   }
+  const incoming = readCriterionLine(line);
+  if (incoming) {
+    const stored = declared.get(
+      identityKey({ kind: incoming.kind, twin: incoming.tag ?? primary, text: incoming.text })
+    );
+    if (stored !== void 0) throw new DuplicateCriterionError(path, line, stored);
+  }
   if (insertAt === -1) {
     insertAt = lines[start + 1]?.trim() === "" ? start + 2 : start + 1;
   }
@@ -4541,7 +4580,7 @@ var DEFAULT_AGENT_COMMAND = `node ${DEFAULT_AGENT_FILE}`;
 var MANIFEST_SCHEMA_URL = "https://pome.sh/schemas/v1/pome.json";
 var MAX_UNREADABLE_PATHS_SHOWN = 5;
 function readPackageVersion() {
-  if ("0.23.31".length > 0) return "0.23.31";
+  if ("0.23.33".length > 0) return "0.23.33";
   try {
     const here = dirname(fileURLToPath(import.meta.url));
     const candidates = [
@@ -4996,7 +5035,7 @@ function createProgram() {
         return;
       }
       {
-        const { runDoctorChecks } = await import('../../checks-QMHVQKS7.js');
+        const { runDoctorChecks } = await import('../../checks-PZSMAVLY.js');
         const { renderDoctorReport } = await import('../../render-ZQQ4UMNO.js');
         const doctorReport = await runDoctorChecks({ mode: useLocal ? "full" : "hosted" });
         if (!doctorReport.ok) {
@@ -5034,7 +5073,7 @@ function createProgram() {
               taskForRuns.config.runs
             );
             if (k > 1) {
-              const { runTrialGroup } = await import('../../runTrialGroup-BMQ2DPSL.js');
+              const { runTrialGroup } = await import('../../runTrialGroup-WXDRCVUE.js');
               const fileForRerun = relative(process.cwd(), file);
               const rerunCommand = defaultTask ? options.trials !== void 0 ? `pome run -n ${k}` : "pome run" : `pome run ${fileForRerun && !fileForRerun.startsWith("..") ? fileForRerun : file} -n ${k}`;
               const groupResult = await runTrialGroup({
@@ -5128,7 +5167,7 @@ function createProgram() {
         process.exitCode = 5;
         return;
       }
-      const { runDemo } = await import('../../runDemo-6WVLS72G.js');
+      const { runDemo } = await import('../../runDemo-EV3CQBDR.js');
       const result = await runDemo({
         apiBase: opts.apiUrl.replace(/\/$/, ""),
         dashboardBase: process.env.POME_DASHBOARD_URL ?? DEFAULT_DASHBOARD_URL,
@@ -5146,7 +5185,7 @@ function createProgram() {
   program.command("doctor").description(
     "Check the agent\u2194twin wiring: pome.json (or pome.yaml) present + valid, the local twin boots + serves, requests routed to the twin (not a hardcoded production host), egress floor active. On failure prints one named cause (file:line where knowable) + one concrete fix and exits non-zero."
   ).action(async () => {
-    const { runDoctorChecks } = await import('../../checks-QMHVQKS7.js');
+    const { runDoctorChecks } = await import('../../checks-PZSMAVLY.js');
     const { renderDoctorReport } = await import('../../render-ZQQ4UMNO.js');
     const report = await runDoctorChecks();
     for (const line of renderDoctorReport(report, { passNote: true })) console.error(line);
@@ -5322,7 +5361,7 @@ function createProgram() {
   ).description(
     "Start a standalone twin as a long-lived foreground server (Ctrl-C to stop)"
   ).action(async (name, options) => {
-    const { runTwinStartCommand } = await import('../../twinStart-FX3UXOCA.js');
+    const { runTwinStartCommand } = await import('../../twinStart-J42TT766.js');
     await runTwinStartCommand(name, options);
   });
   twin.command("reset").argument("[name]", "Twin name (default: github)", "github").description("Reset standalone twin state").action(async (name) => {

package/dist/{src-7F4QNBTF.js → src-A6EH5KVF.js}

@@ -1,4 +1,4 @@
-export { GMAIL_CHECKS } from './chunk-SE44FKWH.js';
+export { GMAIL_CHECKS } from './chunk-KTVFRR56.js';
 import { gmailErrorEnvelope, gmailSeedSchema, parseSeed, defaultSeedState, notFound, invalidArgument, checkFault, unsupported, validateSearchQuery, SEARCH_MAILBOX_MESSAGE_BUDGET, GmailError, parseSearchQuery, CATEGORY_LABELS, parseDate, parseSize, KNOWN_FIELDS, parseDuration, assertHasOperator } from './chunk-NJ246QPJ.js';
 export { DEFAULT_GMAIL_AGENT_EMAIL, GmailError, SEARCH_MAILBOX_MESSAGE_BUDGET, agentPathInboxMailbox, defaultSeedState, gmailErrorEnvelope, gmailSeedSchema, loadSeedFromEnv, parseSearchQuery, parseSeed, validateSearchQuery } from './chunk-NJ246QPJ.js';
 import './chunk-JWJYNAWI.js';

package/dist/twinHarness-XO47SYBH.js

@@ -0,0 +1,5 @@
+export { UnsupportedTwinError, bootTwin } from './chunk-YL4FHI3J.js';
+export { STRIPE_LOCAL_ACCOUNT_ID } from './chunk-3OVV6JUQ.js';
+import './chunk-OW7VVW6X.js';
+import './chunk-VBATFCWR.js';
+import './chunk-SG6ZTIMT.js';

package/dist/{twinStart-FX3UXOCA.js → twinStart-J42TT766.js}

@@ -1,5 +1,5 @@
-import { bootTwin } from './chunk-A46IA7EK.js';
-import { isTwinName, TWIN_NAMES, defaultPortFor, TWIN_REGISTRY } from './chunk-5JJDFIGJ.js';
+import { bootTwin } from './chunk-YL4FHI3J.js';
+import { isTwinName, TWIN_NAMES, defaultPortFor, TWIN_REGISTRY } from './chunk-3OVV6JUQ.js';
 import './chunk-OW7VVW6X.js';
 import './chunk-VBATFCWR.js';
 import './chunk-SG6ZTIMT.js';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pome-sh/cli",
-  "version": "0.23.31",
+  "version": "0.23.33",
   "description": "Digital-twin testing for AI agents \u2014 run tasks against resettable local or hosted twins and record tool-call traces for evaluation on pome.sh.",
   "keywords": [
     "ai",

package/dist/twinHarness-VX3GE7QB.js

@@ -1,5 +0,0 @@
-export { UnsupportedTwinError, bootTwin } from './chunk-A46IA7EK.js';
-export { STRIPE_LOCAL_ACCOUNT_ID } from './chunk-5JJDFIGJ.js';
-import './chunk-OW7VVW6X.js';
-import './chunk-VBATFCWR.js';
-import './chunk-SG6ZTIMT.js';